Cognitive Network Digital Twins: A Survey of Concepts, Architectures, and Future Research Directions

The increasing adoption of data-intensive and real-time Artificial Intelligence (AI) and Internet of Things (IoT) applications in domains such as manufacturing, healthcare, smart cities, and remote mining has placed significant pressure on current 5G and emerging 6G networks, exposing the limitations of conventional reactive network management approaches. In recent years, Network Digital Twins (NDTs) have been proposed to address some of these challenges by creating real-time virtual replicas for monitoring and optimisation. Recent studies have increasingly incorporated learning-based optimisation and adaptive decision-making into Digital Twin-based network systems; however, these capabilities are often developed for specific optimisation or control tasks. In this paper, we define and formalise the concept of Cognitive Network Digital Twins (CNDTs), which extends traditional NDTs with cognitive capabilities including perception, learning, reasoning, prediction, and autonomous decision-making. In emerging data-intensive applications, for example, in Industry 5.0, these autonomous capabilities are intended to complement human expertise and interaction, allowing human operators to provide guidance or intervention where required. We review the state of the art in Digital Twins and intelligent network systems, trace the technological evolution that motivates CNDTs, and introduce a multi-layered architectural taxonomy that organises the key components supporting cognitive functions in CNDTs. We also examine the potential of CNDTs to improve network resilience, efficiency, and scalability and identify key research challenges related to cognitive intelligence, semantic knowledge management, architectural placement, and autonomous control. This work establishes a clear conceptual foundation for CNDTs and highlights their importance as an emerging framework for intelligent and self-adapting network management in future wireless systems.